Eugene Volokh seems to think that John Lott’s responses to his critics are worth reading. I can’t imagine why, except for those interested in abnormal psychology. He has been detected in so many different lies [*], some of them utterly pointless [*], that his words have approximately the net information content of the sound coming from your window fan.
Well, put it down to my interest in abnormal psychology, but I examined both of Lott’s responses. First was a response to Michelle Malkin’s op-ed. This response doesn’t contain anything new, but it does collect together all of his various responses on the survey in the one place, so I have written some comments explaining what is wrong with his responses.
His other response is just his August 20 posting where he finally admitted to making hundreds of coding errors. There were many things wrong with that posting. I already corrected his false implication that the errors only made a difference of thousandths of one percent and also corrected his claim that Ayres and Donohue were keeping their data secret. And the biggest problem was his denial of the fact that correcting his errors eliminated his findings, which I explained here.
However, there was a new bit, in an update hidden in the middle of his posting:
It is also not clear that applying STATA’s clustering command to all counties within a state provides an adequate solution to the problem of cross-correlation. While it is possible that error terms are correlated across jurisdictions within a state, the more important correlations may be among neighboring jurisdictions across state lines. Professors Eric Helland and Alex Tobarrok paper also provides an interesting approach that deals with this problem in an novel way and shows that the results are significant even after these correlations are accounted for.
You know, every now and then Lott comes out with something so breathtakingly dishonest that you have to shake your head at the sheer audacity of it. Helland and Tabarrok show that STATA’s correction for clustering is not enough. That means that after STATA’s correction is made, you still might wrongly decide that your results are significant. But if STATA’s correction means that your results are not significant then Helland and Tabarrok’s further correction will not make them significant. Lott’s postion is like someone arguing that an umbrella will keep you dryer than a raincoat and using that as justification for going out into the rain without a raincoat or an umbrella.
It is true that Helland and Tabarrok found that a couple of results from the original Lott and Mustard paper remained significant after correcting, but none of those results were in Lott’s Confirming “More Guns Less Crime” paper. Correcting Lott’s coding errors causes his results to go away whether you use STATA’s clustering correction (as Lott originally did) or Helland and Tabarrok’s technique.
And Lott still has not responded to the question I have been trying to get him to answer for weeks: “Why did he remove the clustering correction from his model?” Also of interest would be an explanation of why, after I started questioning about his table that used the corrected data, he removed it and replaced it with one based on the miscoded data.